1 / 16

Advanced Compilers CMPSCI 710 Spring 2003 Basic Loop Optimizations

Advanced Compilers CMPSCI 710 Spring 2003 Basic Loop Optimizations. Emery Berger University of Massachusetts, Amherst. Topics. Last time Optimizations using SSA form Constant propagation & dead code elimination Loop invariant code motion This time Loop optimizations Induction variable

amaya-weiss
Télécharger la présentation

Advanced Compilers CMPSCI 710 Spring 2003 Basic Loop Optimizations

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Advanced CompilersCMPSCI 710Spring 2003Basic Loop Optimizations Emery Berger University of Massachusetts, Amherst

  2. Topics • Last time • Optimizations using SSA form • Constant propagation & dead code elimination • Loop invariant code motion • This time • Loop optimizations • Induction variable • Linear test replacement • Loop unrolling • Scalar replacement

  3. Easy Detectionof Loop Induction Variables • Pattern match & check: • Search for “i = i + b” in loop • i is induction variable if no other assignment to i in loop • Pros & Cons: • Easy! • Does not catch all loop induction variablese.g., “i = a * c + 2”

  4. Taxonomy of Induction Variables • basic induction variable: • only definition in loop is assignmentj = j § c, where c is loop invariant • mutual induction variable: • definition is linear function of other induction variable i‘: • i = c1 * i‘§ c2 • i = i‘ / c1 § c2 • family of basic induction variable j: • set of induction variables i such that i is always assigned linear function of j

  5. Strength Reduction • Replace “expensive” op by “cheaper” one • E.g., replace multiply by addition • Apply to induction variable families • Especially: array indexing

  6. Strength Reduction Algorithm • Let i be induction variable in the family of basic induction variable j: • i = c1 * j + c2 • Create new variable i’ • Initialize in pre-header: i’ = c1*j + c2 • Track value of j: after j = j + c3, add i’ = i’ + (c1 * c3)

  7. Strength Reduction Example

  8. Candidates for Strength Reduction • Induction variable IV multiplied by invariant • Recursively: • IV * IV, IV mod constant, IV + IV

  9. Strength Reduction Algorithm

  10. Strength Reduction Examples • basic induction variable: • only definition in loop is assignmentj = j § c, where c is loop invariant • mutual induction variable: • definition is linear function of other induction variable i‘: • i = c1 * i‘§ c2 • i = i‘ / c1 § c2 • family of basic induction variable j: • set of induction variables i such that i is always assigned linear function of j

  11. Linear Test Replacement • Eliminates induction variable! • After strength reduction, loop test is often last use of induction variable • Algorithm: • If only use of IV is loop test and its own increment, and test is always computed • i.e., only one exit from loop • Replace test with equivalent one: • E.g., “i comp k” ) “i_50 comp k*50”

  12. Linear Test Replacement Example

  13. Loop Unrolling • To reduce loop overhead, we can unroll loops • Advantages: • Execute fewer total instructions • More fodder for common subexpression elimination, strength reduction, etc. • Move consecutive access closer together • Disadvantages: • Code bloat • Still updating through memory

  14. Scalar Replacement • Problem: register allocators never keep a[i] in register • Idea: trick allocator • Locate patterns of consistent reuse • Replace load with copy into temporary • Replace store with copy from temporary • May need copies at end of loop • E.g., when reuse spans > 1 iteration • Advantages: • Decreases number of loads and stores • Keeps reused values in registers • Big performance impact (2x, 3x!)

  15. Scalar Replacement Example • Scalar replacement exposes the reuse of a[i] • Traditional scalar analysis – inadequate • Use dependence analysis to understand array references (later)

  16. Next Time • Common Subexpression Elimination • Read ACDI: • Ch. 12, pp. 343-355 • Ch. 13, pp. 378-396

More Related